The CoCounsel alternative.
CoCounsel Legal is the generative tier on top of Westlaw. Casetext and CoCounsel are the same product, since Thomson Reuters acquired Casetext and kept the name.
This page prices the whole deployment rather than the seat, names which models sit in the data path, and says where CoCounsel wins.
Verified 2026-08-24. Every figure carries its source and read date.
Where CoCounsel is still right
Starting here, because the rest is worthless if this part is dishonest.
You already pay for Westlaw
CoCounsel sits on the Westlaw corpus and KeyCite. If that subscription is already signed and the content library is the reason, the AI tier is an incremental decision rather than a platform decision.
Deposition and discovery volume
Document review, deposition preparation, and the Litigation Document Analyzer are mature workflows built against Thomson Reuters content. Marcella does not compete on discovery volume.
One vendor, one invoice
Content, model, and support arriving on a single Thomson Reuters paper has real procurement value at a firm with no IT function. That is a legitimate reason to consolidate.
If you are already on Westlaw, this comparison is narrower than it looks. The question is whether the AI tier is worth the delta, not whether to switch platforms. The Westlaw pricing breakdown is the more useful page.
The cost is five lines, not one
A seat license is one line. Putting a third-party legal AI platform into a law firm has five, and the last one has no price until it lands.
| Cost line | What it is |
|---|---|
| 1. Seat license | Quote-only for most firms. Reported at $150 to $400 per user per month bundled with Westlaw, and $200 to $500 as an AI add-on for firms with an existing subscription. Neither is a Thomson Reuters published rate. |
| 2. Model spend | Agentic workflows consume tokens per run rather than per seat. Long-horizon document work is metered separately from the license on most architectures. |
| 3. Data processor review | A new AI vendor in the data path is a new processor. Approval runs 3 to 6 months at a mid-size firm, and the hours are partner and general counsel hours. |
| 4. Integration and support | Content, model, and orchestration can be different companies. When an output is wrong at 6pm before a filing, the firm has to know which one to call. |
| 5. Custody of the record | Prompts and outputs are electronically stored information. If a third party holds them, the firm is not the custodian of its own work product. No line item. Full explanation. |
Thomson Reuters does not publish a CoCounsel Legal rate. Every figure above is third-party reporting. The one Thomson Reuters number that is published is Westlaw Edge with AI-Assisted Research, from $155.35 per month on a three year term for firms up to ten attorneys.
Which models sit in the data path
Thomson Reuters announced a Model Context Protocol integration connecting Claude to CoCounsel Legal on May 12, 2026, and has said the next generation of CoCounsel Legal is being rebuilt on Anthropic's Claude Agent SDK.
Thomson Reuters also built its own model, Thomson, with a first integration launching August 2026 inside Tabular Analysis in CoCounsel Legal. Depending on the workflow, three parties can sit in a single request: Thomson Reuters content, a Thomson Reuters model, and Anthropic.
That is not a criticism of any of the three. It is a procurement fact. Each one is a data processor your firm approves, and each one is a party that could hold a copy of a prompt.
What Marcella does instead
Inference runs on Azure OpenAI inside Microsoft Azure. The model endpoint has no public network access and is reached only through a private endpoint inside a virtual network, under managed identity rather than an API key. No non-Microsoft model provider receives client content. On Bring Your Own Azure tiers the whole deployment sits in the firm's own tenant, which means no new processor to approve and no new custodian holding the record.
Feature by feature
| Capability | CoCounsel Legal | Marcella |
|---|---|---|
| Underlying corpus | Westlaw content plus KeyCite. The deepest licensed library in the category. | 10M+ judicial opinions from CourtListener and primary sources. Not a licensed treatise publisher. |
| Fabricated citations | Generative. A Stanford RegLab and Yale study reported the Westlaw AI research product near one inaccurate response in three, the weakest of the paid tools tested. | Retrieval only. A citation absent from the index cannot be returned. |
| Models in the data path | Thomson Reuters content, the Thomson model, and Claude via MCP depending on workflow. | Azure OpenAI inside Microsoft Azure. No non-Microsoft provider receives content. |
| New vendor approval | A new AI processor for most firms. 3 to 6 months at mid-size. | Runs in the Microsoft tenant the firm already contracts for. No new processor on BOYA tiers. |
| Pricing transparency | Quote-only. | Published at every tier through BOYA 20. |
| Discovery and deposition workflows | Mature. A genuine product advantage. | Not offered. |
The last row is a CoCounsel win Marcella has no answer for, and for a firm doing high discovery volume it may be the only row that matters.
Questions
Are Casetext and CoCounsel the same product?
Yes. Thomson Reuters acquired Casetext and CoCounsel is the product name that survived. Anyone offering you a comparison of Casetext against CoCounsel is comparing a product to itself.
How much does CoCounsel cost?
Thomson Reuters does not publish a rate for CoCounsel Legal. Third-party reporting puts it at $150 to $400 per user per month bundled with Westlaw, and $200 to $500 per user per month as an AI add-on for firms with an existing Westlaw subscription. Both are estimates.
Which AI models does CoCounsel use?
Thomson Reuters announced a Model Context Protocol integration connecting Claude to CoCounsel Legal on May 12, 2026, and has said the next generation is being rebuilt on Anthropic's Claude Agent SDK. Thomson Reuters also developed its own model, Thomson, with a first integration launching August 2026 inside Tabular Analysis.
Does CoCounsel hallucinate citations?
A Stanford RegLab and Yale study published in 2024 reported that the Westlaw AI-assisted research product produced inaccurate output in roughly one of three responses, the weakest result among the paid tools tested. The failure mode is generative architecture rather than vendor quality.
Can I use Marcella without leaving Westlaw?
Yes. Marcella runs inside Microsoft 365 and does not require removing anything. For firms that need KeyCite and the treatise library, running both is the sensible configuration.
Launch pricing. Runs inside the Microsoft 365 tenant your firm already has, with no new data processor to approve on BOYA tiers.
Join the Waitlist